US Government Orders Shutdown of Anthropic's Fable 5 Over Jailbreak

The US Department of Commerce, led by Secretary Howard Lutnick, has issued an export control directive targeting Anthropic's Fable 5 and Mythos 5 models. The order, delivered at 5:21pm ET on a Friday, classifies the models as subject to national security restrictions, barring access by any foreign national—even Anthropic employees. As the company lacks a system to verify citizenship in real-time, it was forced to abruptly suspend both models for all customers. The justification stems from a narrow jailbreak technique reportedly identified by Amazon, which the government claims could bypass safeguards. Anthropic, however, contests the severity, noting the jailbreak only exposes previously known minor vulnerabilities also present in other publicly available models. In its response, Anthropic emphasized its defense-in-depth strategy, which includes 30-day data retention for monitoring, and stated that no universal jailbreak was found during extensive red-teaming with the US government, UK AISI, and private partners.

The move has sparked debate about the threshold for government intervention in AI model deployment. Critics, including policy researcher Dean W. Ball, describe the decision as 'cartoonish' and based on a misunderstanding of jailbreak mechanics and defense-in-depth. Anthropic had proactively disclosed its risk posture at launch, acknowledging that perfect jailbreak resistance is impossible. The company argues that the specific jailbreak does not enable harmful outcomes beyond what other models already allow. This incident sets a troubling precedent for how regulatory bodies might react to security vulnerabilities in advanced AI systems. With Anthropic forced to shut down its flagship models over a minor exploit, the AI industry faces new uncertainty about export controls being weaponized against specific companies, potentially chilling innovation and deployment of cutting-edge models.